- 📖 Background
- 🚧 Challenges
- 🛠️ Solution Stack
- 💻 Technical Details
- 📊 Key Improvements
- 🧩 Solution Architecture
- 🖥️ Application Walkthrough
- 🎥 Presentation
- 📝 Note
- 💡 Credits
An US based e-commerce platform struggles with its call center data, handling a high volume of customer interactions daily. While it gathers extensive feedback and complaint data, this valuable resource remains underutilized. The challenge lies in analyzing the data to generate insights into customer satisfaction and regional trends, limiting the company's ability to improve services proactively.
- 🧠 Sentiment Analysis: Accurately identifying customer sentiment from call conversations.
- 📝 Call Summarization: Summarizing lengthy calls to extract valuable insights.
- 📍 Regional Insights: Analyzing customer feedback by location to identify trends and concerns.
To address these challenges IBM WatsonX Granite models are levered. Sentiment analysis, call summarization, and feature extraction are done using prompt engineering techniques.
- Python>=3.10
- Name:
NebulaByte/E-Commerce_Customer_Support_Conversations - Source: Hugging Face Dataset
- Column Used:
conversation
- Sentiment Analysis:
ibm/granite-13b-instruct-v2 - Call Summary & Feature Extraction:
ibm/granite-13b-chat-v2
-
Notebook - hackathon-challenge.ipynb
-
Python - hackathon-challenge.py
(Generate API key from IBM Cloud platform)
-
Output HTML Version- hackathon-challenge.html
- 🚀 Boosting Team Productivity: Implementing sentiment analysis & text summarization helps streamline processes, enabling teams to focus on key tasks.
- 🎯 Targeted Training: Analyzing call summaries identifies skill gaps, enabling tailored training initiatives for employees.
- 🏆 Customer Satisfaction-Based HR Incentives: Customer satisfaction metrics derived from call summaries can guide year-end bonuses and promote a customer-centric approach.
- 🔧 Product Improvement Feedback: Recurring negative feedback on products can be flagged, allowing collaboration with vendors to address product quality, delivery, or other issues.
This submission is part of the IBM TechXchange Pre-Conference Watsonx Hackathon.
Refer to the Hackathon Page for more details.
- Anirban Banerjee
- Ajoy Kumar Daga